A Statistical-Feature ML Approach to IP Traffic Classification Based on CUDA

摘要

In modern networks, there exist different applications which generate various different types of network traffic. In order to improve the performance of network management, it is important to identify and classify the internet traffic. The machine learning (ML) technique based on per-flow statistics has been widely used in traffic classification. Different from traditional classification methods, it is insensitive to port number and payload on application level. Our approach in this work is also based on a machine learning method kNN. kNN is a special case of a variable-bandwidth, kernel density "balloon" estimator with a uniform kernel [1]. Although there is no time taken for the construction of the classification model using kNN, it is computationally intensive since it relies on searching neighbor among large sets of d-dimensional vectors. The kNN algorithm may have quite expensive classification steps. CUDA (Compute Unified Device Architecture) is a parallel computing platform and programming model invented by NVIDIA. It enables dramatic increases in computing performance by harnessing the power of the graphics processing unit (GPU) [2]. This paper puts forward a CUDA-based kNN algorithm to classify internet traffic. The experimental results show that the peek speed of traffic classification based on GPU improves greatly compared with that based on CPU. Our approach presents a significant speed improvement through GPU; meanwhile, the results demonstrate the potential applicability of GPU in the field of traffic classification.